Q-学习算法在CGF系统中的应用  

Application of the Q-Learning Algorithm within a CGF System

作　　者：徐如燕[1] 郭齐胜 

机构地区：[1]装甲兵工程学院,北京100072

出　　处：《系统仿真学报》2001年第S2期559-560,563,共3页Journal of System Simulation

基　　金："国家863计划"(No.863-306-ZD10-02-2);"教育部骨干教师资助计划"(No.2001JC09)资助。

摘　　要：CGF中的战场决策仿真十分复杂,要求CGF实体应能根据知识库做出类似人类的决策行为。目前,CGF系统中的决策模块大多是硬编码的,不能满足军用仿真发展的需求,这就要求CGF系统具有学习和自适应能力。Q-学习方法是一种特殊的增强学习方法,可以通过多次迭代计算正/负效益进行主动学习。本文介绍了Q-学习及其在CGF系统中的应用。这种较新的机器学习方法,在CGF中有着良好的应用前景。The simulation of decision-making on the battlefield is complex. The CGF entities should act and react similar to human according to the knowledge base. But up till now, the decision processes in CGF systems are hard-coded. They can hardly meet the requirements of military simulation. More and more CGF systems require the ability of the learning and self-adaptive. Q-learning approach is a specialized type of reinforcement learning that learns actively by determining iteratively the positive and/or negative rewards. In this paper, Q-learning and its application within a CGF system are introduced. Its a promising method to CGF.

关 键 词：决策 Q-学习 增强学习 CGF 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]